Openai Agent Builder Shutdown: What to Do Next

As at 20 October 2026, I'd treat an existing Agent Builder workflow as a migration job. Export the workflow, choose its replacement and test the full business process before 30 November. For owners weighing AI automation for a small business, the immediate decision is where the workflow should run next.

When Does Openai Agent Builder Shut Down?

OpenAI's deprecation page records the announcement on 3 June 2026 and the scheduled shutdown on 30 November 2026. OpenAI does not specify an Australian shutdown time in that notice, so I would plan to finish beforehand.

Agent Builder is OpenAI's visual canvas for multi-step agent workflows. If your enquiry process depends on that canvas, put migration on the work list now. Document the business rules separately so the next move is easier.

What Happens to Chatkit, Evals and Existing Workflows?

ChatKit remains available. That does not preserve an Agent Builder backend indefinitely. OpenAI's ChatKit guide directs new work towards a self-hosted integration with a server-side agent, such as one built with the Agents SDK. The chat front end can stay while its backend changes.

Evals has a separate deadline worth checking: existing evals become read-only on 31 October 2026, and the Evals dashboard and API shut down on 30 November 2026. OpenAI links a migration guide to Promptfoo. If your workflow relies on those evaluations, include them in the migration scope.

Agents Sdk or Chatgpt Workspace Agents: Which Fits?

OpenAI's migration guide separates the options by how you build:

The trade-off is control against the work of maintaining code. OpenAI warns that workflows built around strong determinism may not migrate faithfully to Workspace Agents. Check the options available in your ChatGPT workspace before choosing.

The Agents SDK supports non-OpenAI model providers, too. For a broader shortlist, see my comparison of small-business agent builders.

How Do You Export an Agent Builder Workflow?

Follow OpenAI's export instructions:

1. Open the workflow in Agent Builder.
2. Select Code in the top navigation.
3. Select Agents SDK in the code dialog.
4. Choose TypeScript or Python and copy the complete export.
5. Save a separate description of the workflow's rules, connections and approval points.

The first four steps are OpenAI's export process. The fifth is what I'd add for the business owner. OpenAI explicitly says the process does not convert the workflow graph or guarantee that every behaviour transfers unchanged.

For Workspace Agents, use the export and your written rules to guide rebuilding through natural language. OpenAI requires separate review of connected apps, authentication, publishing and permissions. Keep those details in business-controlled records, following the same principle as avoiding AI vendor lock-in.

How Should an Enquiry Agent Keep Human Approval?

Here's how I'd rebuild an enquiry-handling workflow, as an example design, around five explicit steps:

1. Trigger: receive a website form, chat message or email enquiry.
2. Context: gather the request, service, suburb, relevant CRM history, opening hours and availability.
3. Draft: prepare a reply, suggest a call or site visit and prepare the CRM entry.
4. Approval: hold the reply until a person approves it. Send uncertain or sensitive matters to a person.
5. Record: log the enquiry, approved reply and outcome in the CRM.

The SDK's human-in-the-loop flow can pause sensitive tool calls for approval or rejection. I'd put that gate on the send action and test rejection, missing context and failed CRM writes before launch.

Review the old canvas against OpenAI's node reference: Guardrails, MCP connections, If/else routing and Human approval each need an explicit replacement. Track time to first reply, drafts sent unedited and enquiries becoming booked jobs.

What Does the API Cost, and Where Is Data Stored?

As published in US dollars on OpenAI's pricing page as at October 2026, gpt-5-mini standard short-context usage costs US$0.25 per million input tokens and US$2.00 per million output tokens. These are usage rates, not a complete operating budget. The SDK has an MIT licence; model usage remains billed.

Budget separately for hosting, connections, monitoring and maintenance. I would measure representative enquiries before estimating a monthly bill.

OpenAI's API data documentation, checked 20 October 2026, lists Australian regional storage but no Australian regional processing. Non-US residency requires approval for abuse monitoring controls and a Modified Retention amendment.

The same documentation says API data is not used to train OpenAI models, default abuse-monitoring logs can remain for up to 30 days, and ChatKit thread state stays until deleted. Australian storage does not settle every client-data question. This is general information, not legal advice; check your business's Privacy Act obligations.

Where Does Truespeak Fit?

For an Australian service business that wants enquiry automation designed, built and then run around its existing tools, with a person approving anything sensitive, Truespeak is built for exactly that.

I work hands-on across Claude, GPT, Gemini, Grok and DeepSeek and match the model to the job, so a client does not have to pick a platform first. That matters when a vendor retires a product. As Head of Direct to Consumer at Nikon Australia, I built an AI chatbot that deflected 140+ customer service calls a month. Through managed AI operations, Truespeak monitors systems, fixes failures, reviews exceptions and improves rules after launch. Where a tool has no usable connection, that step stays manual.

The shutdown is also a reminder to budget for automation maintenance after go-live. Keep the workflow rules and approval decisions in records your business controls. If you want help designing that process, start with a discovery call.